| Bispecturm analysis is one of the useful tools for the non-linear systems, non-Gaussian processes and non-minimum phase, and was applied in many fields, such as radar, sonar, physical geography, biomedicine, array processing etc. Underwater Target Recognition is an important area of underwater acoustic signal processing technology of research content, but also one of the difficulties in this area. Feature Extraction is a key link in target classification and recognition, also bispecturm is a growing number of applications in underwater target feature extraction. However, high-dimensional space of the traditional bispecturm feature is not suitable for directly inputting classifier, thus limiting its use. Based on the traditional integral bispectrum theory, these papers, which use a reduced-dimension optimization method, obtained a new local integral bispectrum.This paper studies are as follows:1. The definition, character, algorithm and physical meaning of traditional bispectrum analysis were introduced. The underwater target echo signals were simulated, and the three types of simulation signals on the diagonal slice spectrum and integral bispectrum were calculated.2. The traditional integral bispectrum were optimized by using Fisher discriminant criterion, PCA and singular value decomposition. The nine kinds of optimization of the local integral bispectrum were calculated. Simulation of the signal used nine kinds of new local integration method was calculated. The different methods and different features between the signals were compared.3. Been described the pond experimental overview and calculated the two kinds of goals of suspended and buried with nine kinds of new local integral bispectrum, using BP network classification. A comparative analysis of classification results proved that different feature extraction methods and optimization recognition of the result is not the same, and the conclusions of simulation analysis has been to verify the correctness. |
PDF Full Text Request